Benchling has an opt-in suite of AI-based productivity features. These features use large language models to intelligently interpret and generate natural language and make decisions.
All AI-based features are opt-in and clearly marked in the product, and your data is never used for cross-customer training or training by our AI model providers. For more information, see Data protection and security for AI at Benchling.
Benchling offers various AI agents, features and structure prediction models. AI agents are advanced AI powered capabilities that help you retrieve and capture data on Benchling. AI features provide lighter weight AI functionality to improve productivity. Structure prediction models leverage LLMs to help you visualize protein structures on Benchling.
| Name | Description |
| SQL Writer | SQL Writer helps you write, debug, and modify SQL queries using natural language |
| Notebook Check | Notebook Check flags entry typos, inconsistencies, incomplete sections, and more |
| Plate Annotator |
Plate annotator interprets natural language prompts to assign well roles Note: Plate annotator is in Preview |
| Data Entry Agent | Data Entry Agent supports import of unstructured data into Benchling structured tables by interpreting document context, user instructions and the uploaded file(s) to map data points accurately |
| Deep Research Agent and Ask Agent |
Ask lets you explore your Benchling data quickly through natural-language questions Deep Research answers complex questions by retrieving and analyzing data across your entire Benchling tenant Note: Deep Research with public literature search is in Beta |
| Compose Agent | Compose accelerates notebook entry and template creation through conversational chat |
| Structure prediction models | Structure Prediction Models generate 3D Structures for Amino Acid Sequences on Benchling |
Capabilities marked preview or beta are available, and may be enabled via the tenant admin console using the steps below. Please keep in mind that these features are still a work in progress, and any feedback is welcome via the in-app feedback mechanisms.
Enable Benchling AI
Tenant Admins control Benchling AI capabilities through the Tenant Admin Console, where each capability can be enabled or disabled individually.
To enable Benchling AI capabilities:
- Click on your avatar in the Navigation bar to navigate to the tenant admin console
- Click the Settings tab
- Click AI Settings in the menu
- Identify the feature you would like to enable
- To enable for all users, click Enabled for all users
- To enable for specific users, click Enabled for select users, then use the search box to add all users who will have the feature enabled
- A Settings changed banner will appear, click Save to apply the changes
These settings can be changed at any time by visiting the page again.
Note: Structure prediction models are not currently managed through the Tenant Admin Console. Contact Benchling Support to enable structure prediction models.
Onboard your team
Once your desired Benchling AI capabilities are enabled, share key resources and best practices noted below with your team for them to get started.
Best practices:
- Review our agent specific help center articles to find example use cases or prompts that your team can try
-
Start with high-value, low-risk queries to build confidence
- Example: Instead of a broad question like "Summarize all our cancer research," begin with a focused query such as "Summarize the results from the latest C9ORF72 antibody validation study."
-
Incorporate into recurring workflows (weekly summaries, compliance checks)
- Example: Instead of manually compiling a project update, use a prompt like "Generate a summary of our progress on the lead optimization phase of program X."
-
Keep refining prompts for better accuracy
- Example: A broad prompt like "Summarize our in vivo studies" can be refined to a more specific query: "Summarize the design of in vivo studies ST042 and ST043, including a table of key differences."
-
Save useful queries for reuse
- Example: If a query like "Write a report for how we engineered viral vector X, including the objective, methods, and results" proves effective, save the exact prompt so it can be easily adapted for future reports on other viral vectors.
- Try multiple queries in parallel
Key resources:
- Read about how other teams and organizations are applying Benchling AI
- Check out past Benchling AI trainings and discuss with others in our Community here
Customize Benchling AI usage
Deep Research, Compose, and Notebook Check support custom guidelines which provide additional admin defined context and rules that shape how the relevant AI agents and features work for your tenant.
Custom guidelines are managed by tenant admins in the Tenant Admin Console. To view or modify custom guidelines, open the Tenant Admin console’s settings page following the instructions above and navigate to the AI settings tab on the left hand side.
Custom guidelines can be written in plain language and may contain rules to enforce, areas to focus on, explanations about which rules apply in which situations, etc.
See your Benchling AI usage
Open the Tenant Admin console’s settings page following the instructions above and navigate to the AI Usage tab on the left hand side.
View your tenant's available AI credits and usage data and adjust the date period as needed. If you have questions about Benchling credits and usage, please reach out to Benchling Support.
Benchling Credits
Benchling AI brings AI-powered features, agents, and models directly into the Benchling platform so scientists can accelerate their everyday R&D workflows. Some Benchling AI features are included with your subscription at no additional cost. Agents and models use a credit system — and to ensure every team can start using AI right away, credits are included with every Benchling subscription.
Features that don’t consume credits:
- SQL Writer
- Notebook Checker
Agents and models that do consume credits:
- Ask
- Deep Research
- Compose
- Data Entry Agent
- Structure prediction
The number of credits used per run is fixed by each agent and model, and depends on the complexity of the tasks being performed. See the current Rate Card and learn more here.